Enhancing Continual Learning for Software Vulnerability Prediction: Addressing Catastrophic Forgetting via Hybrid-Confidence-Aware Selective Replay for Temporal LLM Fine-Tuning

Researchers developed Hybrid Class-Aware Selective Replay (Hybrid-CASR), a continual learning method that improves AI-based software vulnerability detection by addressing catastrophic forgetting in temporal scenarios. The method achieved 0.667 Macro-F1 score while reducing training time by 17% compared to baseline approaches on CVE data from 2018-2024.

• →Traditional vulnerability detection models fail in real-world temporal scenarios due to catastrophic forgetting when code bases evolve over time.
• →Hybrid-CASR method outperformed baseline approaches with statistically significant improvements in detecting software vulnerabilities.
• →The approach reduces computational costs by 17% compared to window-only training while maintaining better performance than cumulative training.
• →Research used microsoft/phi-2 with LoRA fine-tuning on CVE-linked datasets spanning six years of vulnerability data.
• →Selective replay with class balancing offers practical accuracy-efficiency trade-offs for continuous AI-based security monitoring.